改进丰富卷积特征算法的液滴边缘检测模型

王慧; 曹召良; 王军

doi:10.37188/CO.2024-0019

改进丰富卷积特征算法的液滴边缘检测模型

doi: 10.37188/CO.2024-0019

王慧^1,,
曹召良^1,,
王军^2, ,

1.
苏州科技大学物理科学与技术学院, 江苏苏州 215009
2.
苏州科技大学电子与信息工程学院, 江苏苏州 215009

基金项目: “十四五”江苏省重点学科资助（No. 2021135）；中国航天科技集团公司第八研究院产学研合作基金资助（No. SAST2020-025）

详细信息

作者简介:
王军（1979—），男，江苏睢宁人，博士，副教授，2005年于中国科学院长春光学精密机械与物理研究所获得博士学位，主要从事光电测控技术、图像信息处理技术以及物联网工程应用方面的研究。E-mail：wjyhl@126.com

中图分类号: TP394.4;TH691.9
计量
- 文章访问数: 89
- HTML全文浏览量: 54
- PDF下载量: 7
- 被引次数: 0
出版历程
- 收稿日期: 2024-01-18
- 修回日期: 2024-01-30
- 录用日期: 2024-03-15
- 网络出版日期: 2024-05-10

Improved droplet edge detection model based on RCF algorithm

WANG Hui^1
,,
CAO Zhao-liang^1
,,
WANG Jun^{2
, ,}

1.
School of Physical Science and Technology, Suzhou University of Science and Technology, Suzhou 215009, China
2.
School of Electronic and Information Engineering, Suzhou University of Science and Technology, Suzhou 215009, China

Funds: Supported by Jiangsu Key Disciplines of the Fourteenth Five-Year Plan (No. 2021135); Industry-University-Institute Cooperation Foundation of the Eighth Research Institute of China Aerospace Science and Technology Corporation (No. SAST2020-025)

More Information

Corresponding author: wjyhl@126.com

摘要

摘要:
液滴图像边缘的高精度提取是测量水接触角较为关键的一环，针对常规边缘提取方法噪声鲁棒性差、边缘提取不完整、精度低的问题，本文提出了一种改进丰富卷积特征（RCF）的液滴边缘检测模型。首先，在深度特征提取阶段引入特征融合模块，使用多个特征让模型更加鲁棒，减少过拟合的风险；其次，设计多感受野模块代替RCF后边的contact层，通过多个感受野来提取更多的语义信息，使边缘细节更加丰富；然后，在模型每一层之前引入高效通道注意力机制，增强模型对图像中重要特征的关注程度；最后，设计并引入MaxBlurPool下采样技术，减少计算量和参数量，提高平移不变性。在自制液滴数据集上的实验结果表明，本文模型的固定轮廓阈值（ODS）提高到0.816、单图像最佳阈值（OIS）提高到0.829、检测准确率高达90.17%，相较原模型提高了1.85个百分点，能够准确检测液滴边缘特征。
- 深度学习 /
- 边缘检测 /
- 水接触角 /
- 特征融合 /
- 曲线拟合
Abstract:
Accurate droplet edge extraction is crucial for measuring water contact angle. To address issues like poor noise robustness, incomplete edge extraction, and low precision in conventional methods, we propose an improved model for droplet edge detection based on Richer Convolutional Feature (RCF) algorithm. Firstly, a feature fusion module is introduced in the deep feature extraction stage to enhance model robustness and reduce overfitting risks. Secondly, a multi-receptive field module replaces the contact layer after RCF to extract more semantic information and enrich edge details. Thirdly, an efficient channel attention mechanism is introduced before each layer of the models to enhance focus on important features of the image. Lastly, the MaxBlurPool downsampling technique is designed and incorporated to reduce computation and parameter requirements while improving translation invariance. Experimental results on a self-made droplet dataset demonstrate that the proposed model achieves an ODS value of 0.816, an OIS value of 0.829, and a detection accuracy of up to 90.17%, which is an improvement of 1.85 percentage points compared to the original model. It can improve accuracy in droplet edge features detections.
- deep learning /
- edge detection /
- water contact angle /
- feature fusion /
- curve fitting

HTML全文

图 1 RCF模型结构

Figure 1. RCF model structure

下载: 全尺寸图片幻灯片

图 2 改进的RCF模型结构

Figure 2. Improved RCF model structure

下载: 全尺寸图片幻灯片

图 3 特征融合

Figure 3. Feature fusion

下载: 全尺寸图片幻灯片

图 4 多感受野模块

Figure 4. Multi-receptive field module

下载: 全尺寸图片幻灯片

图 5 高效注意力机制模块

Figure 5. Efficient attention mechanism module

下载: 全尺寸图片幻灯片

图 6 抗锯齿的最大池化

Figure 6. Anti-aliasing for maximum pooling

下载: 全尺寸图片幻灯片

图 7 一维抗锯齿操作

Figure 7. One-dimensional anti-aliasing operation

下载: 全尺寸图片幻灯片

图 8 数据样本

Figure 8. Data samples

下载: 全尺寸图片幻灯片

图 9 传统RCF损失值与迭代次数的关系

Figure 9. Relationship between the loss and the number of iterations in the conventional RCF algorithm

下载: 全尺寸图片幻灯片

图 10 改进RCF损失值与迭代次数的关系

Figure 10. Relationship between the loss and the number of iterations in the improved RCF algorithm

下载: 全尺寸图片幻灯片

图 11 较小角度液滴边缘检测结果

Figure 11. Edge detection results for small-angle liquid droplet

下载: 全尺寸图片幻灯片

图 12 较大角度液滴边缘检测结果

Figure 12. Edge detection results for large-angle liquid droplet

下载: 全尺寸图片幻灯片

表 1 改进模型各阶段性能

Table 1. Performance of the improved model at each stage

Each stage	Precision/%	Recall/%	F-measure/%
Stage1	72.26	73.38	72.82
Stage2	78.31	81.69	79.96
Stage3	85.65	87.41	86.52
Stage4	88.92	89.15	89.03
Mixed Output	90.17	89.64	89.90

下载: 导出CSV

表 2 本文模型与其他算法结果比较

Table 2. Results comparison of the proposed model and other relevant algorithms

Algorithm	Precision/%	ODS	OIS	OR/%	Time/s
Canny	81.02	0.717	0.642	11.25	0.015
HED	87.49	0.732	0.735	7.36	0.025
RCF	88.32	0.783	0.792	6.74	0.028
Improved-RCF	90.17	0.816	0.829	5.10	0.041

下载: 导出CSV

表 3 消融实验结果对比

Table 3. Comparison of ablation experimental results

Module	ODS	OIS
BasicRCF	0.783	0.792
BasicRCF+FM	0.795	0.812
BasicRCF+FM+DC	0.809	0.817
BasicRCF+FM+DC+EN	0.811	0.823
BasicRCF+FM+DC+EN+MP	0.816	0.829

下载: 导出CSV

参考文献(23)

[1]	周文婷. 仿生超润湿材料的制备及其高粘度原油水分离性能研究[D]. 长春: 吉林大学, 2023. ZHOU W T. Preparation of bionic superwetting material and its high viscosity crude oil-water separation performance[D]. Changchun: Jilin University, 2023. (in Chinese).
[2]	黄云, 黄建超, 肖贵坚, 等. 超疏水表面加工技术及耐磨性能研究进展[J]. 中国机械工程,2024,35(1):2-26. HUANG Y, HUANG J CH, XIAO G J, et al. Research progresses of superhydrophobic surface processing technology and abrasion resistance[J]. China Mechanical Engineering, 2024, 35(1): 2-26. (in Chinese).
[3]	郭伟成, 廖元太, 张洪玉. 润滑水凝胶涂层研究进展[J]. 清华大学学报(自然科学版),2024,64(3):381-392. GUO W CH, LIAO Y T, ZHANG H Y. Research progress in lubricating hydrogel coatings[J]. Journal of Tsinghua University (Science and Technology), 2024, 64(3): 381-392. (in Chinese).
[4]	王晓辉, 李军建, 杨威, 等. 接触角的图像处理与检测[J]. 光电子技术,2011,31(1):14-19. doi: 10.3969/j.issn.1005-488X.2011.01.004 WANG X H, LI J J, YANG W, et al. Measurement on contact angles based on image process[J]. Optoelectronic Technology, 2011, 31(1): 14-19. (in Chinese). doi: 10.3969/j.issn.1005-488X.2011.01.004
[5]	张天, 田汉民, 戎小莹, 等. 粒子群优化Canny算子在高精度接触角测量中的应用研究[J]. 河北工业大学学报,2018,47(3):30-35. ZHANG T, TIAN H M, RONG X Y, et al. Edge detection of Canny operator based on PSO[J]. Journal of Hebei University of Technology, 2018, 47(3): 30-35. (in Chinese).
[6]	ROSENFELD A, THURSTON M. Edge and curve detection for visual scene analysis[J]. IEEE Transactions on Computers, 1971, C-20(5): 562-569. doi: 10.1109/T-C.1971.223290
[7]	MARR D, HILDRETH E. Theory of edge detection[J]. Proceedings of the Royal Society B: Biological Sciences, 1980, 207(1167): 187-217.
[8]	王惠琴, 侯文斌, 黄瑞, 等. 基于深度学习的空间脉冲位置调制多分类检测器[J]. 中国光学(中英文),2023,16(2):415-424. doi: 10.37188/CO.2022-0106 WANG H Q, HOU W B, HUANG R, et al. Spatial pulse position modulation multi-classification detector based on deep learning[J]. Chinese Optics, 2023, 16(2): 415-424. (in Chinese). doi: 10.37188/CO.2022-0106
[9]	张印辉, 庄宏, 何自芬, 等. 氨气泄漏混洗自注意力轻量化红外检测[J]. 中国光学(中英文),2023,16(3):607-619. doi: 10.37188/CO.2022-0127 ZHANG Y H, ZUANG H, HE Z F, et al. Lightweight infrared detection of ammonia leakage using shuffle and self-attention[J]. Chinese Optics, 2023, 16(3): 607-619. (in Chinese). doi: 10.37188/CO.2022-0127
[10]	ARBELÁEZ P, MAIRE M, FOWLKES C, et al. Contour detection and hierarchical image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(5): 898-916. doi: 10.1109/TPAMI.2010.161
[11]	李翠锦, 瞿中. 基于深度学习的图像边缘检测算法综述[J]. 计算机应用,2020,40(11):3280-3288. doi: 10.11772/j.issn.1001-9081.2020030314 LI C J, QU ZH. Review of image edge detection algorithms based on deep learning[J]. Journal of Computer Applications, 2020, 40(11): 3280-3288. (in Chinese). doi: 10.11772/j.issn.1001-9081.2020030314
[12]	REN X F, BO L F. Discriminatively trained sparse code gradients for contour detection[C]. Proceedings of the 25th International Conference on Neural Information Processing Systems, Curran Associates Inc. , 2012: 584-592.
[13]	XIE S N, TU ZH W. Holistically-nested edge detection[J]. International Journal of Computer Vision, 2017, 125(1-3): 3-18. doi: 10.1007/s11263-017-1004-z
[14]	简柯青. 目标物体的轮廓识别关键技术研究[D]. 成都: 电子科技大学, 2022. JIAN K Q. Research on key technologies of object contour detection[D]. Chengdu: University of Electronic Science and Technology of China, 2022. (in Chinese).
[15]	LIU Y, CHENG M M, HU X W, et al. Richer convolutional features for edge detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(8): 1939-1946. doi: 10.1109/TPAMI.2018.2878849
[16]	周聪. 基于立体视觉的光通信器件空间缺陷检测方法研究[D]. 武汉: 华中科技大学, 2022. ZHOU C. Research on spatial defect detection methods of optical communication devices based on stereo vision[D]. Wuhan: Huazhong University of Science and Technology, 2022. (in Chinese).
[17]	朱淑鑫, 周子俊, 顾兴健, 等. 基于RCF网络的遥感图像场景分类研究[J]. 激光与光电子学进展,2021,58(14):1401001. ZHU SH X, ZHOU Z J, GU X J, et al. Scene classification of remote sensing images based on RCF network[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1401001. (in Chinese).
[18]	姜林奇, 宁春玉, 余海涛. 基于多尺度特征与通道特征融合的脑肿瘤良恶性分类模型[J]. 中国光学(中英文),2022,15(6):1339-1349. doi: 10.37188/CO.2022-0067 JIANG L Q, NING CH Y, YU H T. Classification model based on fusion of multi-scale feature and channel feature for benign and malignant brain tumors[J]. Chinese Optics, 2022, 15(6): 1339-1349. (in Chinese). doi: 10.37188/CO.2022-0067
[19]	景年昭, 杨维. 基于RCF的精细边缘检测模型[J]. 计算机应用,2019,39(9):2535-2540. doi: 10.11772/j.issn.1001-9081.2019030462 JING N ZH, YANG W. Fine edge detection model based on RCF[J]. Journal of Computer Applications, 2019, 39(9): 2535-2540. (in Chinese). doi: 10.11772/j.issn.1001-9081.2019030462
[20]	陈浩. 面向非控制环境的自动抠像方法研究[D]. 南昌: 江西师范大学, 2021. CHEN H. Research on automatic keying method for non-controlled environment[D]. Nanchang: Jiangxi Normal University, 2021. (in Chinese).
[21]	黄晨耕. 结合浅层特征与深层特征的光学遥感舰船检测方法研究[D]. 南京: 南京航空航天大学, 2021. HUANG CH G. Research on optical remote sensing ship detection method combining low-level features and high-level features[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2021. (in Chinese).
[22]	张宝, 李小霞, 张婧, 等. 多路感受野引导的特征金字塔小目标检测方法[J]. 制造业自动化,2022,44(11):155-159. doi: 10.3969/j.issn.1009-0134.2022.11.037 ZHANG B, LI X X, ZHANG J, et al. Small target detection method with multiple receptove firlds guided feature pyramid network[J]. Manufacturing Automation, 2022, 44(11): 155-159. (in Chinese). doi: 10.3969/j.issn.1009-0134.2022.11.037
[23]	杨晨, 侯志强, 李新月, 等. 基于CNN-Transformer双模态特征融合的目标检测算法[J]. 光子学报,2024,53(3):0310001. YANG CH, HOU ZH Q, LI X Y, et al. Object detection algorithm based on CNN-transformer dual modal feature fusion[J]. Acta Photonica Sinica, 2024, 53(3): 0310001. (in Chinese).